On This Page

Wording

Note Section 1.3 Reading time: ~5 mins

Exposure Correlation & Double Counting

When explaining why univariate relativities (e.g., Pure Premium Method) suffer from double counting when exposures are correlated:

  • The Issue: Pure premium calculations assume independent distributions. When exposures are correlated, disproportionate concentrations of high-cost or low-cost classes distort the relativities.
  • Suggested Phrasing:

    “Disproportionate exposures are concentrated in high- and low-cost groups. For example, 80% of Class 99 exposures are concentrated in the high-cost Territory Z, whereas the univariate method assumes a balanced distribution. This causes Class 99 to appear costlier than it actually is, resulting in double counting of the territory effect.”


GLM Modeling

  • Signal vs. Noise: Avoid saying GLMs “model noise.”
  • Suggested Phrasing:

    “GLMs may overfit and reflect the noise component in the data in addition to the true underlying signal.”


Claim Process Changes

When describing the impact of a slowdown or speedup in claims processing on development:

  • The Issue: Applying historical development factors to current, changed data levels causes distortion.
  • Suggested Phrasing:

    “Applying historical development factors to current (lower/higher) levels of paid or reported losses will lead to overestimation or underestimation of ultimate claims.”


Predictiveness of Case Reserves

  • Refining Statement: Simplify claims development descriptions.
  • Revision:
    • Draft: “Case outstanding observed are predictive of unobserved case reserves at later maturities.”
    • Better: “Case outstanding observed are predictive of future claim development.”

General Terminology Precision

Be specific in exam responses instead of using generic terms like “results”:

  • Instead of “results”, specify exactly what you mean:
    • Projections (or ultimate estimates)
    • Age-to-age factors (or LDFs)
    • Premium trends or loss ratios
  • Only use “results” when referring to the overall analysis outcomes generally.